Reliability analysis of soil-geogrid pullout models in Japan

The paper examines the accuracy of two geogrid pullout capacity models used in Japan (Public Works Research Center-PWRC, 2000a) by comparing measured capacities from a large database of laboratory pullout tests to predicted capacities. One model is the current default model (Model 1) used when project-specific laboratory pullout testing is not available and the other when this data is available (Model 2). The accuracy of the models is quantified using bias statistics where bias is defined as the ratio of measured pullout capacity to predicted value. Bias statistics are also a necessary precursor for reliability-based load and resistance factor design calibration for the ultimate pullout limit state in the internal stability design of geogrid reinforced soil walls. Bias statistics using Model 2 show that pullout predictions are very accurate with negligible scatter. However, the default model is shown to be very conservative on average with large scatter in bias values which also varies with magnitude of predicted pullout capacity. A modified formulation for Model 1 is proposed that has the same number of empirical coefficients as the current expression (i.e. two). The modified formulation gives improved accuracy based on the computed mean and coefficient of variation of bias values, and eliminates the significant model bias that is present for the current model. A final outcome from analysis of all available pullout data is that a factor of safety F=2 is recommended for allowable stress design (ASD) when using the current PWRC default Model 1, and a value of F=1.25 is recommended when using Model 2 with project-specific pullout testing.